---
pretty_name: df_eval
license: other
task_categories:
- audio-classification
language:
- en
- zh
tags:
- deepfake
- speech
- anti-spoofing
- audio
size_categories:
- 1M<n<10M
configs:
- config_name: add_2022_track_1
  data_files:
  - split: test
    path: add_2022_track_1/test-*
- config_name: add_2022_track_3
  data_files:
  - split: test
    path: add_2022_track_3/test-*
- config_name: add_2023_round_1
  data_files:
  - split: test
    path: add_2023_round_1/test-*
- config_name: add_2023_round_2
  data_files:
  - split: test
    path: add_2023_round_2/test-*
- config_name: asvspoof_2019
  data_files:
  - split: test
    path: asvspoof_2019/test-*
- config_name: asvspoof_2021_df
  data_files:
  - split: test
    path: asvspoof_2021_df/test-*
- config_name: asvspoof_2021_la
  data_files:
  - split: test
    path: asvspoof_2021_la/test-*
- config_name: asvspoof_2024
  data_files:
  - split: test
    path: asvspoof_2024/test-*
- config_name: codecfake
  data_files:
  - split: test
    path: codecfake/test-*
- config_name: dfadd
  data_files:
  - split: test
    path: dfadd/test-*
- config_name: fake_or_real
  data_files:
  - split: test
    path: fake_or_real/test-*
- config_name: in_the_wild
  data_files:
  - split: test
    path: in_the_wild/test-*
- config_name: librisevoc
  data_files:
  - split: test
    path: librisevoc/test-*
- config_name: sonar
  data_files:
  - split: test
    path: sonar/test-*
- config_name: speechfake
  data_files:
  - split: test
    path: speechfake/test-*
- config_name: speechfake_en
  data_files:
  - split: test
    path: speechfake_en/test-*
---

# df_eval

Public **evaluation-only** speech deepfake detection dataset, organized like Common Voice language configs:
each **config** is a standard eval protocol (ASVspoof, ADD, In-the-Wild, …) with embedded audio.

Companion code: [github.com/Shuo-H/df_eval](https://github.com/Shuo-H/df_eval)

> Configs are added as uploads complete. Declared configs below match currently available parquet shards on the Hub.

## Load

```python
from datasets import load_dataset

ds = load_dataset("shuohann/df_eval", name="sonar", split="test")
print(ds[0]["audio"], ds[0]["label"], ds[0]["file_id"])
```

## Schema

| column   | type                         | description                          |
|----------|------------------------------|--------------------------------------|
| audio    | Audio                        | waveform                             |
| label    | ClassLabel(bonafide, spoof)  | ground-truth                         |
| file_id  | string                       | basename of original file            |
| source   | string                       | config name                          |
| path     | string                       | original protocol path (metadata)    |

Split: **test** only.

## Planned configs

Aligned with [Speech DF Arena](https://huggingface.co/spaces/Speech-Arena-2025/Speech-DF-Arena) protocol files:

- `asvspoof_2019`, `asvspoof_2021_la`, `asvspoof_2021_df`, `asvspoof_2024`
- `add_2022_track_1`, `add_2022_track_3`, `add_2023_round_1`, `add_2023_round_2`
- `codecfake`, `dfadd`, `fake_or_real`, `in_the_wild`, `librisevoc`, `sonar`

## License / redistribution notice

This Hub dataset **repackages** evaluation audio from multiple third-party corpora.
Each source has its **own license** (often research / non-commercial). Upstream sources include:

- [ASVspoof 2019](https://zenodo.org/records/6906306)
- [ASVspoof 2021 LA/DF](https://zenodo.org/records/4837263)
- [ASVspoof 2024](https://zenodo.org/records/14498691)
- [ADD 2022 / 2023](https://zenodo.org/)
- [FakeOrReal](https://bil.eecs.yorku.ca/share/for-norm.tar.gz)
- [Codecfake](https://zenodo.org/records/11169781)
- [DFADD](https://github.com/isjwdu/DFADD)
- [LibriSeVoc](https://zenodo.org/records/15127251)
- [SONAR](https://drive.google.com/file/d/1UBfih4BUDA29rmDtPhUibX-5Xqy2XNlQ/view)
- [In The Wild](https://deepfake-total.com/in_the_wild)

**You are responsible for complying with each original license.** Prefer citing and downloading from the original authors when required. The uploader does not claim ownership of the underlying audio.

## Citation

Please cite the original dataset papers for any config you use, and Speech DF Arena if you use their protocol definitions.